the good the bad and the lucky ceo pay and skillweb-docs.stern.nyu.edu/clb/docs/06-005.pdfharvard...

THE GOOD, THE BAD AND THE LUCKY: CEO PAY AND SKILL*

ROBERT DAINES VINAY B NAIR

LEWIS KORNHAUSER**

FIRST VERSION: NOVEMBER 2004 THIS VERSION: AUGUST 2005

* We thank Bill Allen, Heitor Almeida, John Core, Yakov Amihud, Martijn Cremers, Ron Gilson, Kose John, Paul Oyer, Roberta Romano, Antoinette Schoar, Morten Sorensen and Geoff Tate, as well as seminar participants at Berkeley, British Columbia, University of Alberta, Stanford, the NYU-Wharton Conference on Finance and Law, the Asian Institute of Corporate Governance at Seoul, Korea, the American Law and Economics Association meetings, the European Finance Association meetings in Moscow and the conference on corporate governance at the Swedish Institute of Financial Research for helpful comments and discussions; and Jinghua Yan for excellent research assistance. ** Daines is from Stanford Law School, Nair is from the Wharton School at the University of Pennsylvania and Kornhauser is from the New York University Law School.

1

THE GOOD, THE BAD AND THE LUCKY: CEO PAY AND SKILL

Robert Daines Vinay B. Nair

Lewis Kornhauser

Abstract

CEO compensation varies widely, even within industries. In this paper, we investigate whether differences in skill explain these differences in CEO pay. Using the idea that skilled CEOs should be more likely to continue prior good performance and more likely to reverse prior poor performance, we develop a new methodology to detect whether skill is related to pay. We find that highly paid CEOs are more skilled than their less well paid peers when pay is performance-based and when there is a large shareholder. This detected link between pay and skill is strong even when we examine industry-wide declines: highly paid CEOs are more likely to reverse the firm’s fortunes. We also examine CEO turnovers and show that the firm’s post-turnover performance is related to differences between the two CEO’s pay levels. These results highlight conditions where pay and skill are linked, and hence identify firms where high pay appears to have no justification.

2

One of the great, as-yet-unsolved problems in the country today is executive compensation and how it is determined. Securities and Exchange Commission chairman, William H. Donaldson1

Is it a problem of bad apples, or is it the barrel? Harvard Business School Dean, Kim Clark2

PART I. INTRODUCTION

Very few business topics are as hotly contested as the salaries of CEOs of public firms.

The amount that CEOs are paid and the structure of their pay is frequently debated in the popular

press, television programs, proposed legislation, political campaigns, magazine cover stories and

academic research. Outrage over CEO pay has forced important changes at major firms; Jack

Welch of GE was forced to give back part of his pay when investors complained it was excessive

and Richard Grasso of the NYSE was forced to resign altogether when the details of his pay were

revealed.

One obvious reason for the interest in CEO pay is its striking increase. In 1992, the

average CEO of an S&P 500 firm earned $2.7 million. By its peak in 2000, average pay for these

CEOs had increased to over $14 million – an increase of more than 400%. The increase in CEO

pay is more striking in relative terms. Twelve years ago, CEOs at major U.S. corporations were

paid 82 times the average earnings of a blue collar worker; last year they were paid more than 400

times the average blue collar worker. This huge increase in executive compensation has been

especially controversial because CEOs are sometimes paid large sums even as the firm’s results

1 National Press Club speech, August 2003 2 National Press Club speech, February 2003

3

deteriorate; CEOs at WorldCom, Tyco, and Enron, collected over $100 million on average in the

year prior to the scandals at or collapse of their firms.

These facts and spectacular governance failures at important firms have caused many to

conclude that the process for setting CEO pay and, more generally, the governance of public firms

is badly broken. Critics such as Bebchuk and Fried (2004) conclude that CEOs are overpaid

because they have too much influence over the board that should be monitoring them. Thus, CEOs

are effectively able set their own pay and the CEO’s compensation contract is badly distorted. In

this view, CEO pay is the product of badly functioning corporate governance. Such arguments

also suggest that cases of excessive CEO pay reflect a systematic social problem of “fat-cat”

CEOs skimming money at shareholders’ expense and therefore a systematic breakdown of

governance, thus demonstrating the need for widespread governance reform.

Others are more sanguine, arguing that the process for determining CEO pay is not

systematically broken. Holmstrom and Kaplan (2003) argue that US corporate governance works

relatively well and that any problems with CEO pay have not erased the comparative advantages

of the U.S. system.3 Murphy (2002) argues that CEO pay levels and practices are generally

consistent with good governance and that distortions from optimal contracting reflect the

perceived impact of accounting and tax rules. In this view, compensation problems reflect

breakdowns in particular firms, but do not indicate a general problem in compensation or in the

governance of public firms generally.

Previous research seeking to distinguish between these competing views of CEO pay

typically examines whether a CEO’s pay changes with contemporaneous changes in firm

3Reasons other than governance failures may explain the increase in pay unrelated to performance. For example, Murphy and Zabojnik (2004), Oyer (2004), and Himmelberg and Hubbard (2000) argue that CEO pay reflects changes in labor market for CEOs. Bolton, Scheinkman and Xiong. (2003) argue that it may even be in shareholders’ interest to avoid incentive pay if capital markets are speculative.

4

performance. Agency theory suggests that the CEO pay should be linked to changes in firm value

in order to align managers’ interests with shareholders.4 To the degree pay doesn’t change with

performance, deviations are taken as evidence of CEO skimming excessive compensation.

Surprisingly, however, there is little evidence on the basic question: is CEO pay related to

CEO skill? Are highly paid CEOs better than their more poorly paid peers? Given the potential

agency problems that affect CEO pay, it is difficult to know how well the market for CEO pay

functions. However, an indirect indicator of a well functioning CEO labor market would be that

the ranking of CEOs based on their pay levels should reflect their skill levels.

The question of whether CEO pay and skill are related is not only important in evaluating

the appropriateness of CEO pay levels but also in evaluating the governance of public firms. If

pay and skill are related, high salaries will not necessarily be evidence that “fat cat” CEOs capture

the board. Just as sports teams may decide to pay high salaries to attract or retain valuable players,

boards may also pay more for especially talented managers. If, on the other hand, pay and skill

are unrelated, the process for setting CEO pay is likely to be badly broken and highly paid CEOs

may be over-paid, regardless of which model describes the process of setting executive

compensation. The question of CEO pay and skill is thus indirect evidence of the quality of

governance in public firms.

To examine whether pay and skill are linked, we introduce a new methodology. The

intuition behind our methodology is straightforward: firms run by good CEOs should consistently

do better than firms run by bad CEOs, taking industry factors into account. If a firm has recently

performed poorly relative to its peers, a skilled CEO will be more likely to reverse the firm’s

4 There are several papers that investigate factors, such as risk, that affect the optimal amount of incentive pay. See Murphy (1999) and Core, Guay and Larcker (2001) for a survey of this literature.

5

fortunes, while a less skilled CEO will be more likely to continue the poor performance.5 If, on

the other hand, a firm has performed well relative to its peers, a skilled CEO will consistently be

more likely to continue the good performance, while a less skilled CEO will increase the chance of

a bad outcome. Thus, good CEOs will reverse poor performance and continue positive

performance. “Bad” CEOs continue poor performance and reverse positive performance.

“Lucky” CEOs are highly paid, but perform no differently from their lower-paid (and “unlucky”)

peers.

Our measure for skill – the persistence of positive performance and the reversal of poor

performance -- is a modification of the performance persistence measure used to evaluate skill

among mutual fund managers (Brown and Goetzmann, 1995), hedge fund managers (Brown,

Goetzmann and Ibbotson, 1999) and investment analysts (Mikhail, Walther and Willis, 2004) and

has several advantages for studying industrial firms and CEOs. First, by conditioning on past

performance, we try to ensure that we are comparing CEOs to other CEOs in a similar position. A

firm’s opportunities may be a function of its past performance, so by controlling for past

performance, we avoid comparing CEOs with whatever different prospects the firms might have

as a result of their prior performance.

Second, this method allows us to avoid some of the econometric problems that have

affected prior research. One reason that prior research has not previously examined the

relationship between CEO pay and CEO skill is that it is difficult to separate CEO effects from

firm effects. Bertrand and Schoar (2003) deal with this by examining CEOs who move from one

firm to another. This methodology helps them to distinguish the styles of particular CEOs, but for

much of the analysis restricts the sample to roughly 117 CEOs (out of around 500 top executives)

5 We give a formal definition in Part II.

6

that switched firms. However, CEOs who switch firms may be different from CEOs who do not

change firms. The potential difference between CEOs who move and those who do not move

might be important in that the design of CEO pay is said to be a wide-spread problem and to be

particularly bad in firms when the CEO is entrenched and thus hasn’t moved. While our

methodology does not explicitly separate the effects of individual CEOs, it does reduce such

concerns (as discussed in Part II, any identified skill will likely be related to the CEO) and can be

used to shed light on how firms respond to industry shocks. Moreover, our method allows us to

consider a large sample of public firms, which is useful given the widespread nature of the

potential problem of executive compensation. We can also distinguish between “good,” “bad” and

“lucky” CEOs rather than simply examining whether there is persistence in general.

Our empirical strategy is straightforward. If pay and skill are related, current

compensation will be related to future performance; if CEO pay levels are unrelated to CEO skill,

current compensation will be unrelated to future performance. High skill would yield high current

pay, and the following year, the executive’s skill would lead to good firm performance. Thus, if

skill and pay are related, one should find that highly paid CEOs are more likely to continue good

performance and to reverse poor performance. If pay and skill are unrelated, one should find no

such relation.6

Using industry-adjusted returns on assets as our measure of firm performance, we find

evidence that highly paid CEOs are more skilled when the CEO receives performance-based pay.

In the presence of performance-based pay, higher paid CEOs are more likely to continue good

performance and to reverse poor-performance than their more poorly paid peers. Further, this link

between skill and pay exists only when there is an external blockholder to monitor management,

6 This relationship should hold in a functioning labor market where a CEO’s reservation wages and pay level are increasing with a CEO’s outside opportunities, skill or marginal product.

7

consistent with the idea that performance-based pay works best when blockholders monitor

(Mulinaithan and Bertrand, 2001). While our focus is on return on assets as a measure of firm

performance, we also investigate positive persistence and reversals in equity returns. Using

annual abnormal returns, likely to be associated with market surprises, we find some consistent

evidence that highly paid CEOs with high performance-based pay are more likely to be skilled.7

We also examine firm responses to a common industry shock and find evidence that highly

paid CEOs show superior skill in responding to industry downturns when pay is highly

performance-based, but only mixed evidence that pay and skill are connected when pay is not

performance-based. We also find that a new CEO who gets a higher pay than the departing CEO

is more likely to reverse prior poor performance, relative to CEOs who are paid similarly or lower

than the departing CEO, if the pay package is performance-based. Strikingly, if the highly paid

new CEO’s pay is not performance-based, the CEO is more likely to continue prior poor

performance. Finally, we also document that the link between pay and skill appears to be stronger

in industries where pay variation is higher.

Our results indicate that a class of highly paid CEOs exhibits superior skill (those with

high performance-based pay and an external blockholder) and another class (CEOs without an

external blockholder or high performance-based pay) shows no such evidence of superior skill.

Thus there appears no justification for the relatively high pay of CEOs without an external

blockholder and low performance-based pay. While the results suggests that such CEOs may be

overpaid, we are unable to take a stand on whether the pay associated with the skilled CEOs is

‘excessive’. Although, the relative pay levels appear reasonable in such cases – that is, that the

7 Why these abnormal returns should persist is not clear given that CEO pay is public. A possible reason might be the relatively short time period used, during which the market does not sufficiently update its estimate of CEO skill.

8

highly paid appear to be better than the poorly paid -- it might still be the case that all CEOs are

overpaid.

Part II describes our methodology and the data. Part III describes our findings on the

connection between pay and skill, the importance of performance-based pay and the role of

blockholders. Part IV investigates the role of firm size and uses equity returns as a performance

measure. Part V examines the impact of an individual CEO by examining the effect of CEO

turnover and industry-wide trends. Part VI examines the association between pay and skill as a

function of pay dispersion within an industry. Part VII concludes.

PART II. METHODOLOGY AND DATA

A. CEO SKILL

A class of models of the impact of the CEO on firm performance underlies our estimations.

Basically, we consider models in which firm performance x depends on the state of nature � and

CEO skill �: x = f(�,�). For purposes of illustration, we suppress the state of nature and treat firm

performance as a random variable that has a density g(x;�) (and cumulative distribution function

G(x;�)). A minimal conception of CEO skill implies that a skilled CEO improves firm

performance stochastically. Specifically, we require that,

Definition 1: If two CEOs A and A’ have skills � and �’ respectively such that � > �’,

then A is more skilled than A’ if and only if the distribution G(x;�) of firm returns

generated by A first order stochastically dominates the distribution G(x;�’) of firm returns

generated by A’.

9

Notice that, in our formalization of skill, a skilled CEO must do more than merely improve

the mean performance of the firm over a less skilled CEO. First-order stochastic dominance not

only implies that the mean of G(x;�) is greater than the mean of G(x;�’); it also implies that the

distribution of returns generated by the skilled CEO is shifted to the right of the distribution of

returns generated by the less skilled CEO: if � > �’ , then, for all x, G(x;�’) � G(x;�).

More generally, we might condition the distribution of firm returns on more than CEO

skill. For example, firm returns in period t might, for example, depend not only on CEO skill �t in

period t but also on firm performance xt-1 in period t-1. Now we simply require that if � > �’ as

before G(xt; xt-1, �) first order stochastically dominates G(xt; xt-1, �’). For any fixed level of

performance in t-1, the probability that the more skilled CEO will produce returns greater than xt

(for any xt) is greater than the probability that the less skilled CEO will produce returns greater

than xt.

Specifically, then, our definition of a skilled CEO has two implications that we will take as

jointly necessary for the empirical manifestation of skill. First, a skilled CEO is more likely to

continue good performance than a less skilled CEO. Second, a skilled CEO is more likely to

reverse bad performance than a less skilled CEO.

We note that skill itself is a function of CEO effort and CEO talent. Neither effort nor

talent is directly observable but economic theory predicts that CEO effort will depend on the

structure of CEO pay. In this article, we seek to determine whether CEO skill and CEO pay are

related. Given our definition of skill, our approach to this question follows immediately: if pay

and skill are related, current compensation will be related to future performance. If CEO pay

10

levels are unrelated to CEO skill, then current compensation will be unrelated to future

performance.

It is worth noting at the outset that our approach differs from the vast majority of research

on CEO pay. Most prior research examines the structure and design of CEO pay or, in other

words, the relationship between pay and contemporaneous performance.8 By contrast, we

examine pay and its relation to future firm performance. Thus, while the prior literature typically

uses firm performance as the independent variable and the CEO’s pay as the dependent variable,

we use prior CEO pay as an independent variable and future firm performance as the dependent

variable.

B. EMPIRICAL METHODOLOGY

Since we are interested in detecting if pay levels are related to CEO skill, we estimate the

following regression for two categories of CEO Pay (high pay and low pay):

Perfi,t = a + a1 IBAD + b (Perfi,t-1*IGOOD) + c (Perfi,t-1*IBAD).

where Perfi,t is the measure of the firm’s performance, and IBAD (IGOOD) is an indicator variable

equal to one if the firm’s performance was worse (better) than the industry median. We are

specifically interested in whether the coefficients for performance persistence (b) and reversals (c)

differ with different CEO pay levels.

Alternatively, we estimate both regressions jointly as follows

8 An important exception, discussed in Part V, is Hayes and Schaefer (2000). If boards observe and reward a CEO’s contribution before the market does, then regressing current pay on future performance will not capture CEO skill: current high pay and future good performance might both just reflect good current firm performance and good current performance may not be related to CEO skill, but luck.

11

Perfi,t = a + b HighPay + c IBAD + d (IBAD* HighPayi,t-1) + e (Perfi,t-1*IGOOD) + f (Perfi,t-1*IBAD) +

P * HighPayi,t-1*(Perfi,t-1*IGOOD) + R * HighPayi,t-1*(Perfi,t-1*IBAD) + ε

where HighPayt-1 is a dummy variable set equal to 1 if the CEO pay was in the top 40% of the

industry the prior year, equal to 0 if the CEO pay was in the bottom 40% of the industry, and

otherwise set to missing.9 The highly paid CEO is more likely to continue prior good performance

if P is positive.10 If highly paid CEOs are also highly skilled, in addition to P, the estimated

coefficient R will be significant as well. However R will be negative (rather than positive)

because we adjust firm performance for the industry median and hence firm performance is

actually negative in bad years.11 Because a skilled CEO is more likely to turn the firm around and

produce positive results, a skilled CEO should have a negative coefficient on the prior (negative)

bad performance.12

Since two firms that have performed differently in the past might have very different

possibilities in the future, we condition on prior performance being above industry median or

below industry median levels. This ensures that we compare CEOs managing firms with similar

prior results. Conditioning on prior performance has another advantage. There may be firm

specific factors that increase both pay and persistence in prior performance. For instance, it might

9 We are interested in whether CEO pay and skill are connected. Therefore, we examine the CEO's total compensation and do not adjust our calculations for other factors that might explain excess CEO compensation such as firm size, characteristics or CEO education. See, e.g., Core, Holthausen and Larker (1999). Such other factors may affect the CEO's skill, but we are agnostic as to the source of CEO skill and are interested in whether skill and pay are related. We thus do not examine excess compensation. If factors such as age and education are directly related to skill, controlling for all these factors would remove any variation in CEO pay and prevent us from examining whether overall pay and skill are related. 10 P represents the performance of highly-paid CEOs relative to poorly paid CEOs following years the firm had good performance and R represents the performance of highly-paid CEOs relative to poorly paid CEOs following years the firm had bad performance. 11 To adjust for industry performance, we subtract the industry median from each firm’s performance. 12 Results are consistent when we estimate the regression with two-year lags, rather than one year lags.

12

be the case that large firms have both a higher pay and, due to inertia, have greater overall

performance persistence. We mitigate – but do not completely remove - such concerns by

defining skill as the persistence of superior performance and the reversal of poor performance (or

turnarounds).13 If we find only one or the other, we do not conclude that pay and skill are related.

In terms of the regression, we therefore look for both a significant positive coefficient for highly

paid CEOs given prior good performance (P) and a significant negative coefficient for highly paid

CEOs given prior bad performance (R). This pattern would indicate that, compared with poorly

paid CEOs, highly paid CEOs are both more likely to continue good performance and more likely

to reverse prior bad performance. We would interpret that as evidence that skill and pay are

linked. The alternative interpretation would be that some firm factor (unrelated to the CEO’s

actions and skill) is correlated with high pay, reversals of bad performance and persistence of

good performance. We are unable to think of plausible candidate for such a factor unrelated to

CEO skill.14

B. DATA

We first compute CEO pay, defined as the sum of the cash, bonus payments, option grants

and stock grants. We then compute industry adjusted pay-levels for all CEOs by subtracting the

median CEO pay in the same industry, using the 48 industry classification provided by Fama-

French (1997). We then sort the CEOs in pay-quintiles based on industry adjusted overall pay and 13 Momentum in stock prices will not explain any observed skill. It is true that stock price momentum might lead to performance persistence for CEOs whose compensation were related to the firm’s stock price, however we define skill as performance persistence when the firm’s prior performance was good, and reversals (rather than momentum) if the prior performance was bad. This pattern of reversals will be inconsistent with explanations based on momentum. Also note that any survivorship bias will only explain observed skill if the most highly paid CEOs are more likely to fail and thus disappear from the sample, thus exaggerating the good performance the next year. This seems unlikely. 14 One potential objection is that, by conditioning on prior poor performance, we might identify someone as skilled when they have only done well relative to poor performers, which is too low a standard for calling someone skilled. However, because we condition both on good and bad performance, highly paid CEOs will need to do well consistently (following good years and following bad years) to be identified as skilled.

13

classify CEOs as highly paid if they are in the top 2 pay quintiles for their industry (the top

40%).15

We use a firm’s industry-adjusted return on assets as our primary measure of firm

performance.16 As a secondary measure, we use annual stock returns. To the degree stock market

participants accurately estimate the CEO’s skill, we would not expect skilled CEOs to be able to

regularly produce positive abnormal returns. Nevertheless, for robustness purposes, we also

present some results using the firm’s stock price performance allowing for the possibility that a

skilled CEO ‘surprises’ the market. Thus, we include in our sample all firms for which we can

find stock return data on CRSP, accounting data on Compustat, and compensation data on

Execucomp for the years 1992-2001. This produces a sample of 12,569 firm years, in which there

are 2,373 unique firms and 3,579 unique CEOs.

In regressions reported below, we exclude firms if there is a turnover either that year or in

the previous year. This ensures that we do not use the performance of one CEO to estimate the

performance of another. The resulting sample thus avoids years involving (or following) CEO

turnovers and has 2,284 firms and 2,880 CEOs (10,043 firm years). The sample of firm-years

involving CEO turnovers, as used in section V, contains 980 firms. Finally, we also obtained data,

from Thomson’s CDA Spectrum database, to identify firms with external blockholders, defined as

institutional investors who hold at least 5% of the firm’s stock.17

Table 1 below provides summary data for our sample. Panel A lists descriptive statistics.

Mean CEO pay in our sample is 4.39 million, of which 46% is characterized as performance-

15 Results are robust to defining high pay as those CEOs in the top 20% and are not presented here. 16 See discussion in Part V on why biases associated with accounting data are less likely to generate the results we document. 17 See Bertrand and Mullainathan (2001) for the role of blockholders in the design of CEO pay and Cremers and Nair (2005) for the impact of blockholders on firm valuation and equity returns.

14

based.18 The top five spots are occupied by Charles Wang of Computer Associates ($1.3 billion in

1998), Steve Jobs of Apple ($600 million in 2000), Gregory Reyes of Brocade ($370 million in

2001) and Thomas Siebel of Siebel ($293 million in 2000, and $244 million in 2001). On the

other end of the scale were 18 CEOs who received no pay.

Variation in CEO pay is large, with a standard deviation in our sample of roughly $18

million. Figure 1a provides detailed information by industry. The difference between CEOs paid

at the 90th percentile and those paid at the 10th percentile ranges from $1.66 million in Fabricated

Products and Machinery and $2.62 million in Textiles (on the low side) to $24.2 million in

Finance and nearly $25 million in Telecom (on the high side). As evident in Figure 1b, the

standard deviation in CEO pay is closely connected to the industry average pay. Note that there is

very little between industry variation in the fraction of CEO pay that is performance-based. Even

as average pay increases, the incentive fraction remains relatively flat at between 35% to 50% of

total CEO pay.

Panel B of Table 1 reports correlation coefficients between the main variables of interest.

CEO pay is positively correlated with firm size and with the fraction that is performance-based, as

expected. Our proxy for shareholder monitoring of CEO performance, the existence of a 5%

blockholder, is negatively correlated with total CEO pay and positively correlated with the fraction

that is performance-based. As an initial matter, we also find in unreported regressions that high

pay and the firm’s future ROA are positively related. The effect is statistically significant at the

1% level and large. A highly paid CEO is associated with an increase in industry adjusted ROA of

1.04%.

[Insert Table 1 and Figure 1 about here]

18 For a measure of incentive pay, we use ((total pay– salary–bonus)/total pay).

15

PART III THE LINK BETWEEN CEO PAY AND SKILL

We first estimate a pooled cross-sectional regression estimating performance persistence

and reversals for all firms in our sample. The estimated regression coefficients are reported in

Table 2.19 We find that, when compared to their poorly paid peers, highly paid CEOs are

significantly more likely to exhibit consistently good performance. The estimated coefficient is

0.09 and significant at the 5% level. However, we find no evidence that highly paid CEOs are

more likely to reverse poor performance (estimated coefficient is -0.01, but not statistically

significant). Since we define CEO skill as production of greater positive persistence and the

increased likelihood of reversing poor performance, we cannot say whether CEO pay and skill are

generally linked. Not surprisingly, we also find performance persistence generally.

A. PERFORMANCE-BASED PAY

As discussed in Part II, CEO skill can be viewed as a combination of the executive’s innate

ability (or talent) and the CEO’s effort. Two CEOs with different talent endowments but who

exert no effort are more likely to perform similarly than are CEOs who exert different amounts of

effort. Since higher performance-based pay should induce higher effort, we would expect to find a

stronger link between pay and skill when the CEO has greater incentives to exert effort.20 Thus,

we would observe a greater link between pay and skill when pay is tied to performance. We

19 The statistical significance of the results presented in the paper is robust to clustering of standard errors by firm. 20 By investigating the link between pay and skill in categories based on level of performance-based pay, we, however, compare high paid CEOs with high performance-based pay to low paid CEOs with high performance-based pay. We thus emphasize on the impact of talent when effort incentives are high.

16

therefore examine the link between pay and skill when performance-based pay (stock and option-

based compensation) constitutes an important part of the CEO’s total pay package.

We first divide the sample into firms lead by CEOs with high (or low) performance-based

pay, depending on whether the CEO’s fraction of stock and option based compensation is above

(below) the median level in the same industry. We then separately run the same regression for

both groups21

Consistent with the prediction that pay and skill are more closely linked in the presence of

high-powered incentives, we find greater evidence of skill for highly paid CEOs where pay is

performance-based. As reported in Table 2, panel B below, when CEO pay is performance-based,

highly paid CEOs were more likely to continue superior performance and also more likely to

reverse inferior performance than their less richly compensated peers. Conditional on good prior

performance, the estimated coefficient for highly paid CEOs was 0.41 and significant at the 1%

level. Conditional on poor prior performance, the estimated coefficient was -0.39 and also

significant at the 1% level. Since the persistence and reversal coefficients for low paid CEOs is

0.30 and 0.75 respectively, high paid CEOs are more than twice more likely to continue prior good

performance and reverse prior poor performance. By contrast, absent performance-based pay, we

find no evidence of differential skill for highly paid CEOs; there is no significant difference in the

likelihood of a reversal of poor performance. See Table 2, panel C.

This effect is economically meaningful. The impact of a highly paid CEO on firm

performance can be roughly estimated using these data. The difference between the likelihood of

persistence between the high paid and the low paid CEOs is 0.41 following good performance (see 21 Analyzing performance pay also sheds light on a potential interpretation that all agents have same reservation wages but different contracts are offered. More skilled agents choose pay structures with greater incentive pay and hence also get a higher pay level. We think this is unlikely to be the complete story, since there are many CEOs with high pay but not high incentive pay. Moreover, even within an industry, where firms are similar, there are differences in incentive pay and pay levels.

17

Table 2, Panel B). For an average firm in this category (median firm size = 12.9 bn), the average

ROA conditional on good performance is 4.36%. Thus the increase in earnings associated with a

highly paid CEO is $211 million (.41 x 4.36% x 12.9 bn). A similar calculation for the marginal

impact of a highly paid CEOs following years the firm has performed poorly yields an estimate of

$112 million in additional earnings.

Thus, we find that highly paid CEOs with performance-based pay are significantly more

likely to reverse bad performance and to continue good performance than their more poorly paid

peers. By contrast, for CEOs with low powered incentives, there is no such evidence that high

CEO pay is associated with superior results. These results are consistent with the idea that CEOs

have differing abilities and that pay and skill are indeed linked when the pay is performance-

based.

[Insert Table 2 and 3 about here]

B. MONITORING SHAREHOLDERS AND PERFORMANCE-BASED PAY

We next examine a refinement of these results on performance-based pay. Bertrand and

Mullainathan (2001) have shown that stock and option based pay are not likely to be related to

CEO effort in the absence of monitoring shareholders. Absent such monitoring shareholders, high

performance-based pay seems to compensate CEOs for luck rather than effort. Thus performance-

based pay should perhaps instead be characterized as a combination of high stock and option

based pay and an external blockholder. We therefore redo our results on performance-based pay

for firms with and without a blockholder, where blockholder is defined as a non-managerial

investor holding at least 5% of the stock.

18

Generally consistent with Bertrand and Mullainathan (2001), the results suggest that

incentives work better in the presence of a blockholder. As reported in Panel Table 3, highly paid

CEOs with a blockholder and high stock and option-based pay are both more likely to continue

good performance (estimated coefficient 0.18, significant at 10%) and to reverse poor performance

(estimated coefficient is -0.77, significant at 1%) than are their less well-paid counterparts.

However, absent a blockholder, highly paid CEOs are no more likely to continue positive

performance than their more poorly paid peers, regardless of the pay for performance intensity of

the CEO’s pay. CEOs in firms with high incentives but without a blockholder do appear to

reverse prior bad performance (estimated coefficient -0.37, significant at 1%)

The effect of a highly paid CEO in these firms is also economically meaningful. The

difference between the likelihood of reversals between the high paid and the low paid CEOs is -

0.77 following poor performance (see Table 3). For an average firm in this category (median firm

size=10.52 bn), the average ROA conditional on poor performance is -1.6%. Thus, the increase in

earnings associated with a highly paid CEO is $129 million (-0.77 x -1.6% x 10.52). A similar

calculation for the impact of a highly paid CEOs when the firm has performed well yields $77

million.

This evidence is thus consistent with the idea that CEOs pay and skill are related in firms

with both performance-based pay and with monitoring shareholders. There is some mixed

evidence that pay and skill are related when the firm has either a monitoring shareholder or

performance-based pay, but not both. Absent both a monitoring shareholder and performance-

based pay, there is no evidence that CEO pay and skill are related. A caveat is that there is a

relatively few number of firms without institutional blockholders and the lack of statistical

significance for firms without blockholders may be related to the lower number of observations.

19

Part IV. EXTENSIONS

A. HOW SIGNIFICANT ARE THE REVERSALS?

We earlier reported that firms were significantly more likely to turn around poor

performance when managed by a highly-paid CEO, so long as the pay was performance-based

pay. But how much did things improve? A negative estimated coefficient (R) could indicate that

following the bad year, either: (a) firm performance improved, but that it was still below industry

median (relative improvement); or (b) the firm had improved sufficiently that it was now

performing better than the median firm (absolute improvement).

We investigate this question by estimating the probability that a firm will do badly, again

conditional on whether the firm’s prior performance was poor. We estimate a logistic regression

for the form:

Poort = C + Poort-1 + Highpayt-1 + Highpay x Poort-1

where HighPay reflects highly paid CEOs as defined earlier (top 40% of industry) and poor is a

dummy variable that takes the value of 1 if firm performance was below industry median.

As evidenced in Table 4, we find that firms managed by highly paid CEOs are less likely

to underperform their industry the following year. Additionally, supportive of our earlier results,

we find that this negative relation between CEO pay levels and future poor performance is

stronger when pay is performance based. When pay is performance based the coefficient on the

Highpay dummy is -0.39 and is significant at the 1% level. However, when performance based

component in pay is low, the coefficient on Highpay is -0.26 and is significant at only the 10%

level.

20

[Insert Tables 4 and 5 here]

B. IMPORTANCE OF FIRM SIZE

In our basic regression testing the overall link between pay and skill, we found a link

between pay levels and positive persistence, but no evidence that pay and skill were linked

generally. One reason for that may be that there are firm factors associated with both high pay and

persistence. One such firm characteristic could be firm size. Large firms might have both more

inertia in performance and higher CEO pay; large firms, like large ships, may take time to turn

around, making it difficult to find evidence of turnaround skill in large firms. We therefore

examine the link between pay and skill for large and small firms. Note that this hypothesized

inertia would not explain why highly paid CEOs are more likely to reverse poor performance,

though it may explain why we do not find reversals in some instances. Firms are classified as

large if their asset size is greater than the median asset size in their industry. We now investigate

if pay and skill are related in large and small firms. In these regressions, we use both one year and

two year lags between performance measures to allow for the possibility that large firms take

longer to turn around.

Using one-year lags, we find that the earlier documented positive persistence is much

stronger in small firms (0.16 compared to 0.09 for the entire sample). The estimated coefficient

for reversals for highly paid CEOs of small firms is negative, but only marginally significant (p

value of 0.13). Thus, the inertia of relatively large firms does not cause the overall positive

persistence we observe in the sample. It is also perhaps tempting to conclude that the link between

pay and skill is looser in large firms. On the other hand it might be the case that large firms, on an

average, take longer to turn around. Consequently, we redo our regressions using two year lags

21

between performance measures. We now observe a strikingly higher likelihood of reversals in

large firms run by high paid CEOs and although the estimated relationship between prior good

performance and subsequent performance is positive, the significance is weak. Thus, taking a

conservative view, the link between pay and skill appears to exist for both small and large firms

but appears stronger in small firms.

C. USING EQUITY RETURNS AS A MEASURE OF PERFORMANCE

We also examine the relationship between high-CEO pay and abnormal returns. As noted

earlier, stock returns are another potential measure of firm performance, however they are more

problematic in that to the degree that the market anticipates the CEO’s skill, risk-adjusted stock

returns will not vary for firms run by CEOs of differing skills. Nevertheless, we allow for the

possibility that a skilled CEO can surprise the market and examine the relationship between CEO

pay and subsequent realized stock returns.

We compute the yearly abnormal return for each firm in the sample. The abnormal return

is computed relative to the Fama-French four factor model that includes the market factor, size

factor, book to market factor and the momentum factor. Both of these measures of performance

are computed relative to the firm’s industry, where industry adjustments are done using the 48

industries classified by Fama and French.22 To reduce the noise in these yearly abnormal returns

we (1) use daily returns and (2) form quintiles based on a firm’s measure of abnormal returns,

making it less likely that our results are driven solely by noise in measures of return.23 We now

22 The abnormal return to a value weighted industry portfolio is subtracted from the abnormal return of each individual firm in the same industry to adjust for industry returns. 23 Our results are qualitatively similar, but statistically weaker, when we use the abnormal returns directly instead of quintiles and are omitted here. In results below, we find similar results when we use an ordered logistic regression rather than an ordinary least squares regression.

22

perform persistence and reversal regressions using the industry-adjusted abnormal return as a

measure of performance.

The estimated regression coefficients are reported in Table 6. Results using abnormal

returns as a measure of performance are similar to those using return on assets, with one notable

difference. Compared to their poorly paid peers, highly paid CEOs have a significant positive

persistence coefficient. The estimated persistence coefficient is 0.11 and significant at the 5%

level. However, in contrast to the return on assets results, the dummy on high CEO pay in this

case is negative and significant (-0.15, significant at the 1% level). Thus, the difference in the

likelihood of persistence associated with high paid CEOs is –0.15 + 0.11 x prior good

performance. Consequently, the greater likelihood of positive persistence outweighs the negative

effect on high pay only for extreme cases of good performance.24 Thus there seems to be some

evidence that high paid CEOs show some skill if firm performance is very good (> 80th percentile)

and not otherwise. These results, on the one hand, suggest that pay and skill are related precisely

when firm performance is furthest away from the industry median and hence is less likely to be

due to small changes around industry median and, on the other hand, suggest that the link between

pay and skill using abnormal returns is weaker than with accounting measures and depends on

existing firm performance.

We also investigate if the link between pay and skill varies with performance-based pay.

As with return on assets, we find that highly paid CEOs are more likely to continue prior poor

performance if the pay consists of a high performance-based component. Interestingly, in such

cases, we also find a weakly significant negative coefficient on reversals (p-value of 0.10). When

high pay has a low performance-based component, we find no such effects. While confirming our

24 Prior good performance takes the value of 1 or 2, since we look at industry-adjusted quintile measures.

23

results on the importance of performance-based pay and thus providing a robustness check, it is

also worth noting that both these effects only hold in cases of extreme prior good and bad

performances.


PART V CEO SKILL, INDUSTRY DOWNTURNS AND CEO TURNOVER

We have seen that pay and skill are correlated when CEO-pay is performance-based and

when there is a large shareholder to monitor management. In this section, we examine two

additional tests that shed light on the question of whether this effect is attributable to something

specific about the CEO’s skills.

A. CEO SKILL IN INDUSTRY-WIDE DOWNTURNS

We have argued that skilled CEOs are more likely to continue prior good performance and

more likely to reverse prior poor performance. However, the reversal of poor performance implies

that the firm recently did poorly. In order to reconcile this variable performance with a more stable

notion of CEO skill, we posit that this poor performance could be due to industry level shocks.

However, it might then be natural to ask how better skilled CEOs react to such industry level

changes. We consequently investigate how highly paid CEOs respond to performance

deteriorating changes in the industry.

24

If highly-paid CEOs respond better to industry-wide downturns, this would bolster our finding

that the CEO skill played some role in reversing a widespread trend. Since it is difficult to

completely identify exogenous shocks for a large number of industries, we instead identify those

instances where the average industry performance declined from one year to the next. If we

continued to observe a link between pay and skill following industry-wide secular declines, this

would be support for the argument that highly-paid CEOs are more likely to withstand adverse

industry trends and thus more skilled, even if they sometimes have poor outcomes. Our test to

identify performance persistence and reversals is the same as before, but is done only for those

industry-years where the average industry ROA has declined. It is useful to note that the average

industry performance does not affect the firm performance measures since these firm performance

measures are industry adjusted.

As shown in Table 7, given performance-based pay, firms run by highly paid CEOs are

significantly more likely to perform well, both when the firm had previously performed well

relative to its peers (good performance) and when it had performed poorly relative to its peers (bad

performance). Here the persistence coefficient is particularly interesting suggesting that better

performing CEOs remain better performing CEOs even in conditions when the entire industry has

performed poorly. Further, both the persistence and reversal coefficients are significant at the 1%

level and again economically meaningful (estimated coefficients were 0.65 for the persistence of

positive performance and -0.39 for the reversal following a bad year). By contrast, when we

examine firms where CEO pay is below the median level of performance-based pay, highly paid

CEOs do not show consistent evidence of skill. Highly paid CEOs are then more likely to reverse

negative performance (estimated coefficient of -0.18, significant at 1%) but are no more likely to

continue prior positive performance.

25

Thus, we find evidence consistent with the earlier results that CEO pay and skill are linked

when performance-based pay is high, even when we limit the sample to firms experiencing an

industry wide decline in performance.

B. CEO SKILL AND CEO TURNOVER

We next test if skill and pay are related by examining a firm’s performance persistence in a

sample associated with CEO turnover. Changes in CEOs could be associated with changes in

CEO skill. If pay is indeed related to skill, pay levels should be different when skill varies.

Consequently, we would expect that a new CEO who is paid differently from the old CEO would

perform differently as well.

We begin by testing whether changes in CEO pay are associated with changes in

performance persistence, without conditioning on prior performance. If CEO pay and skill are

connected, CEOs with similar wages should be similarly skilled and perform similarly. Greater

differences in pay suggest greater differences in skill and therefore less correlation between the

firm’s performance under the old CEO and its performance under the new CEO. In each year t,

we select firms that witnessed a turnover the previous year (t-1). We then relate the firm’s

performance in year prior to the old CEO’s departure (t-2) to the firm’s performance after the new

CEO has had a year on the job (t).

As before, each CEO’s pay is categorized into 5 categories, ranging from 1 to 5, based on

quintiles formed using industry pay levels. The difference between the two CEO’s pay is the

measured simply as the difference between their quintiles. We compute the absolute value of

26

these changes, so the change in compensation varies from 0 to 4.25 All observations with

compensation change greater than 2 are lumped together to form three categories of compensation

change of 0, 1, and >2. We then estimate the firm’s performance persistence for each of these

three categories.

As reported in Table 8 below, we find that firm’s post-turnover performance is less similar

to its pre-turnover performance if the new CEO’s pay is not similar to the former CEO’s pay.

This is consistent with the idea that changes in pay are connected to changes in skill, but is only

suggestive. Ideally we would examine this sub sample in greater detail to see whether pay

increases are associated with performance improvements, and how performance-based pay affect

pay and skill.26 However, we have only around 200 observations with pay increases and a similar

number for pay reductions. Further partitions, based on firm characteristics, reduce our chances of

finding statistical significance. There is also a potential bias with the CEO turnover sample that

makes it difficult to detect positive persistence. CEO turnovers are often associated with poor

performance and there are fewer instances of CEO turnover following good performance. Thus it

is difficult to observe any statistically significant positive persistence following good performance,

but we have a better chance in detecting patterns among reversals of poor performance.

Nevertheless, we perform these regressions based on persistence and reversals and

unsurprisingly find that most of our estimated coefficients are statistically insignificant. See Table

9 below. However, we find that an increase in CEO pay is associated with a greater likelihood of

reversing poor performance only when pay is highly dependent on firm performance. These

25 The new CEO’s compensation information is from year t-1 and the old CEOs compensation information is from date t-3. 26 Note that by considering positive persistence and reversals we are already partitioning the sample into two categories based on prior performance. Any further partitioning (e.g., on incentive pay) generates atleast four categories.

27

findings confirm our earlier findings on the importance of performance-based pay in the link

between pay and skill.


These findings are also of interest because they address an alternative interpretation of our

earlier results using the sample without CEO turnovers. Hayes and Schaeffer (2000) show that the

portion of CEO pay that is unexplained by current performance might reflect the fact that boards

observe performance that public markets will only observe in the future. That is, boards might

pay CEOs today for their good performance that will yield improved results tomorrow. If higher

unexplained variation in pay is also related to higher CEO pay levels, then it might be the case that

high pay is correlated with future good performance because boards observe and pay for CEO

effort not yet reflected in public measures of performance.27 However, in a sample where there is

a CEO change, there is no history for the new CEO and therefore any link between pay and skill is

unlikely to be driven by realized good performance that is not yet reflected in public measures of

performance. The fact that the performance change across CEOs depends on the difference in

CEO pay is not likely to be driven by the fact that boards may pay CEOs for performance that is

not yet reflected in public measures of performance.28

27 Note that higher unexplained variation is not always related to higher pay. For example, a CEO who gets a high pay concurrently with high firm performance has little unexplained variation in CEO pay but a CEO who gets a high pay when firm performance is poor has a high unexplained variation. Thus, the interpretation that pay reflects realized performance not yet reflected in public performance measures is more applicable for reversals, where both pay is high and unexplained variation is high too. Encouragingly, it is exactly here too that the turnover results suggest that pay and skill are related. 28 It is also useful to address a concern that Hayes and Schaefer (2000) acknowledge in the interpretation of their results. They discuss the possibility that the link between higher unexplained compensation and high future performance might be due to earnings manipulation by the CEO. If managers indeed smooth earnings between time periods and are paid for the ‘true’ current earnings, it should be difficult to detect reversals because managers would “borrow” from future good times to boost current bad times. Thus, smoothing would mask rather than explain any reversals.

28

Nevertheless, if boards do pay CEOs in advance of firm performance, this might produce a

pattern of high pay and subsequently better performance even if firm performance varies randomly

from year to year. To rule out the possibility that CEO skill is simply an artifact of this sort of

process, we first examine the frequency of pay changes. If CEO skill is not constant over time, as

this interpretation suggests, then CEO pay will vary a lot over time as well. We examine

compensation changes from year to year (in terms of industry adjusted quintile) and find that in

fact more than 90% of the time there is either no change in CEO pay or a change of 1 quintile.

Thus, it appears that pay is relatively stable over time and does not change a lot from year to year

as this alternative would suggest. We also re-ran our regressions using only CEOs that were

consistently identified as highly-paid in consecutive years, with largely similar results.

PART VI. PAY DISPERSION AND CEO SKILL We began the paper by noting the widely varying degree of pay dispersion within

industries - as shown in Table 2, the intra-industry difference between CEOs paid at the 90th

percentile and those paid at the 10th percentile ranges from $1.66 million in Fabricated Products

and Machinery to nearly $25 million in Telecom. Are these differing levels of pay dispersion due

to differing importance of CEO skill?

To answer this, we examine the association between CEO pay and skill at different levels

of this pay dispersion. When CEO pay differences are relatively compressed, it may mean that

CEO skill is relatively less important in determining firm performance in that industry. Then, the

link between pay and skill would be weaker in such industries. We therefore calculate the

difference between the 90th percentile CEO pay and the 10th percentile CEO pay in each industry

and divide the sample into above and below median pay-dispersion. For each category, we then

estimate the regressions that test whether pay and skill are related. Results are in Table 10.

29

In industries with below-median pay dispersion, there is no evidence that highly paid

CEOs are more likely to turnaround negative performance or to continue positive performance.

Even the signs on the estimated persistence and reversal coefficients are of contrary to what one

would expect if pay and skill were indeed linked. However, in industries with dispersion above the

median, there is some evidence. Highly paid CEOs in these industries are more likely to continue

positive performance. Although they are not more likely to turnaround negative performance, the

reversal coefficient is negative as well. This again is suggestive of a difference between highly

paid and less well paid CEOs in industries where pay dispersion is large. This finding is then

consistent with the view that CEO skill is important in these industries and hence differences in

CEO skill are associated with differences in pay.

PART VII. CONCLUSION

Our analysis has two important conclusions, one methodological and one substantive. We

have introduced a new measure of CEO skill that allows us to assess the relation of CEO skill and

pay. We assess skill in terms of the CEO’s role in perpetuating good firm performance and in

reversing or turning around poor firm performance. Substantively, we provide evidence that

highly paid CEOs are likely to be associated with superior subsequent firm performance,

particularly when the CEO receives performance-based pay and when a large shareholder

monitors CEO performance. These effects are both statistically significant and economically

important, suggesting that highly paid CEOs are in fact “good” where pay is performance-based

and in the presence of a blockholder. By contrast, where pay is not performance-based, we find no

evidence that highly paid CEOs are in fact more skilled at improving the firm’s profitability,

suggesting that highly paid CEOs in these environments may just be “lucky.” We observe a

connection between pay and skill even in firms experiencing an industry-wide secular decline.

30

These findings shed light on a highly debated question - ‘are CEOs overpaid?” On the one

hand, we find a set of firms in which highly paid CEOs perform better (i.e. firms where the CEO

receives performance-based pay or where a large shareholder monitors CEO performance). In

these firms, highly paid CEOs are actually on average “good” – they consistently do better than

their peers. We conclude that compensation problems in such firms are likely outliers (“a few bad

apples”) rather than signs of a systematic problem. We find no evidence that pay and skill are

connected in firms where CEO pay is relatively insensitive to firm performance or when there is

no shareholder blockholder to monitor management – we refer to these highly paid CEOs as

“lucky” given that there is no evidence of superior performance to justify the higher pay.

31

REFERENCES Bebchuk, Lucian Arye and Jesse M. Fried, Pay without Performance, (Cambridge, MA: MIT Press, 2004). Bertrand, Marianne and Antoinette Schoar, “Managing with Style: The Effect of Managers on Firm

Policies,” The Quarterly Journal of Economics, CXVIII (2003), 4. Bertrand, Marianne and Sendhil Mullainathan, “Are CEO’s Rewarded for Luck? The Ones Without

Principals Are,” Quarterly Journal of Economics, CXVI (2001), 901-932. Bolton, Patrick, Jose Scheinkman and Wei Xiong, “Executive Compensation and Short Termist Behavior

in Speculative Markets”, NBER Working Paper No. W9722 (2003). Brown, Stephen and William Goetzmann, “Performance Persistence,” Journal of Finance, L (1995), 679-

698. Brown, Stephen, William Goetzmann and Roger Ibbotson, “Offshore Hedge Funds: Survival and

Performance, 1989-95,” The Journal of Business, LXXII (1999), 91-117. Carhart, Mark, “On Persistence in Mutual Fund Performance,” Journal of Finance, LII (1997), 57–82. Core, John E., Wayne Guay, and David F. Larcker, “Executive Equity Compensation and Incentives: A

Survey,” Working Paper, Wharton School (2001). Core, John E., Robert W. Holthausen, and David F. Larcker, “Corporate Governance, Chief Executive

Officer Compensation, and Firm Performance,” Journal of Financial Economics, LI (1999), 371-406.

Fama, E. F., and K. R. French, “Industry Costs of Equity,” Journal of Financial Economics, XCIII (1997),

153- 194. Hayes, Rachel M. and Scott Schaefer, “Implicit Contracts and the Explanatory Power of Top Executive

Compensation for Future Performance,” RAND Journal of Economics, XXXI (2000), 273-293. Himmelberg, Charles and R. Glen Hubbard, “Incentive Pay and the Market for CEOs: An Analysis of Pay-

for-Performance Sensitivity," Working Paper (2000). Holmstrom, Bengt and Steven Kaplan, “The State of U.S. Corporate Governance: What’s Right and

What’s Wrong?” ECGI Working Paper (2003). Mikhail, M., B. Walther and R. Willis, “Do Security Analysts Exhibit Persistent Differences in Stock

Picking Ability?” Journal of Financial Economics, LXXIV (2004), 67-91. Murphy, Kevin J. and Jan Zobonjik, “CEO Pay and Appointments: A Market Based Explanation for

Recent Trends,” American Economic Review, XCIV (2004), 192-196. Murphy, Kevin J. 1999. “Executive Compensation.” in Handbook of Labor Economics. Orley Ashenfelter

and David Card, eds. North Holland, pp. 2485-2563.

32

Murphy, Kevin J., “Explaining Executive Compensation: Managerial Power vs. the Perceived Cost of Stock Options,” University of Chicago Law Review, LXIX (2002), 847-869.

Oyer, Paul, “Why Do Firms Use Incentives That Have No Incentive Effects?” Journal of Finance, LIX

(2004), 1619-1650.

33

TABLE 1 DESCRIPTIVE STATISTICS

This table reports descriptive statistics of a sample of 2,412 firms (12,725 firm years) between 1992-2001. Blockholder is a dummy variable indicating whether there is a shareholder holding at least 5% of the firm’s stock. The sample includes all firms for which executive compensation, accounting and stock price data is available on Execucomp, Compustat and CRSP. Panel A gives summary statistics and Panel B lists correlation coefficients, significant at 5% (**) and 1% (***) levels. Panel A

PAY (1,000)

PERFORMANCE-BASED PAY

ASSETS (1,000,000)

BLOCK ADJ.ROA ADJ.EQ.RETURN

Mean 4,439 0.46 8,738 0.75 1.96 -0.03 Stand. Dev. 16,731 0.29 36,998 0.42 11.55 0.22 N 12,666 12,640 12,692 10,147 11319 12725

Panel B Performance-based pay

0.23***

Assets 0.02*** 0.11***

Block -0.02** 0.04**

- 0.11***

Adj.roa 0.03*** 0.02*** -0.02*** -0.02***

Adj.eq.return -0.00 0.00 0.00 -0.02* 0.10***

34

FIGURE 1

CEO pay by industry10th percentile, median and 90th percentile

0

5000

10000

15000

20000

25000

30000

35000

FabP

rTx

tls Util

Agr

icG

old

Rub

brS

teel

Per

Sv

LabE

qC

oal

Pap

erM

eals

Box

esC

hem

sM

ach

Clth

sW

hlsl

Gun

sB

ooks

Med

Eq

Tran

sE

lcE

qO

ilTo

ysB

ldM

tR

tail

Shi

psA

utos

Cns

trH

shld

Chi

psS

oda

Hlth

Dru

gsFo

odIn

sur

Aer

oB

usS

vB

anks

Com

psM

ines

Fun

Sm

oke

Bee

rFi

nTe

lcm

0

5000

10000

15000

20000

25000

30000

35000

40000

45000

50000

FabP

rTx

tls Util

Agr

icG

old

Rub

brS

teel

PerS

vLa

bEq

Coa

lP

aper

Mea

lsB

oxes

Che

ms

Mac

hC

lths

Whl

slG

uns

Book

sM

edE

qTr

ans

ElcE

q Oil

Toys

BldM

tR

tail

Ship

sAu

tos

Cns

trH

shld

Chi

psSo

daH

lthD

rugs

Food

Insu

rA

ero

BusS

vBa

nks

Com

psM

ines Fun

Smok

eB

eer

Fin

Telc

m

Standard Deviation

Average Pay

Fraction incentive pay

Table 2 CEO PAY AND SKILL

This table reports regression estimates of the performance of public firms between 1993 and 2001. The dependent variable is the firm’s industry-adjusted return on assets. Independent variables include a dummy variable for whether the firm was managed by a highly paid CEOs (defined as those CEOs whose pay is in the top 40% of industry adjusted pay in the prior year). This dummy variable estimates the effect of highly paid CEOs relative to poorly paid CEOs (those CEOs whose industry adjusted pay is in the lower 40% in the prior year). The regression also includes a control for the prior year’s ROA. These variables are all interacted with dummy variable indicating whether the performance was above the industry median (good prior performance) or below the industry median (poor prior performance). To ease exposition, we report the estimated coefficients in two columns. The sample includes all firms for which executive compensation, accounting and stock price data is available on Execucomp, Compustat and CRSP. Standard errors are in parentheses and are adjusted for clustering. Estimated coefficients are significant at 1(***), 5(**), and 10 (*) percent.

A B C

All firms

Above Median Performance-Based Pay

Below Median Performance-Based Pay

Good prior performance

Bad prior performance





Prior performance for highly paid CEOs

0.09** (0.04)

-0.01 (0.02)

0.41*** 0.07

-0.39*** 0.09

0.19*** 0.07

0.04 0.04

Prior performance 0.66** (0.03)

0.35*** (0.02)

0.30*** 0.06

0.65*** 0.09

0.73*** 0.03

0.33*** 0.02

Highly paid CEO dummy -0.23 (0.36)

0.05 (0.57)

-0.68 0.65

-0.86 1.05

-0.77 0.66

0.06 1.11

Intercept 0.75*** (0.27)

-1.43*** (0.42)

1.38** 0.60

-0.22* 0.96

0.50 0.32

-1.47*** 0.49

R-squared 0.31 0.31 0.33

N 5,808 2,987 2,809

TABLE 3 MANAGERIAL SKILL BY PREFORMANCED BASED PAY AND BLOCKHOLDER

We report association between a highly paid CEO and firm performance. Firms are first partitioned into high performance-based pay (defined as those where the fraction of CEO pay tied to stock price is above the industry median) or low performance-based pay (below the industry median). We then further partition firms by the existence of an institutional shareholder that owns at least 5% of the firm’s shares. The table reports results from a regression estimating the firm’s industry-adjusted accounting performance. We report only the estimated coefficients for firms lead by highly paid CEOs conditional on positive prior performance (above industry median) and prior bad (below industry median) performance (or “reversals”), that estimates the effect of highly paid CEOs relative to poorly paid CEOs. The regression model includes other unreported independent variables, including a constant, a dummy for highly paid CEOs, and controls for prior year performance, and interaction terms. The sample includes all firms for which executive compensation, accounting and stock price data is available on Execucomp, Compustat and CRSP. Estimated coefficients are significant at 1(***), 5(**), and 10 (*) percent.

Above median performance - based pay

Below median performance - based pay





Blockholder

0.18* (0.11)

n=1,872

- 0.77*** (0.10)

0.18** (0.09)

n=1,705

- 0.04 (0.06)

No Blockholder 0.04 (0.12) n=593

- 0.37*** (0.12)

0.15 (0.13) n=519

- 0.62 (1.12)

37

TABLE 4 MANAGERIAL SKILL AND THE LIKELIHOOD OF BAD PERFORMANCE

This table reports the results of logit regressions estimating the performance of public firms between 1993 and 2001. The dependent variable is an indicator variable set equal to 1 if the firm’s industry-adjusted return on assets is negative (i.e. below industry median). Independent variables include a dummy variable for whether the firm was managed by a highly paid CEOs (defined as those CEOs whose pay is in the top 40% of industry adjusted pay in the prior year). This dummy variable estimates the effect of highly paid CEOs relative to poorly paid CEOs (those CEOs whose industry adjusted pay is in the lower 40% in the prior year). Other variables include an indicator variable for whether the firm performed poorly (i.e. below industry median) in the prior year, and an interaction term. The sample includes all firms for which executive compensation, accounting and stock price data is available on Execucomp, Compustat and CRSP. Standard errors are in parentheses and are adjusted for clustering. Estimated coefficients are significant at 1(***), 5(**), and 10 (*) percent.


Above median Performance-Based Pay

Highly paid CEO dummy - 0.26 * (0.15)

- 0.39*** (0.15)

Prior poor performance 2.79*** (0.09)

2.05*** (0.20)

Highly paid CEO dummy * Prior poor performance

0.01 (0.23)

0.31 (0.22)

Intercept - 1.54*** (0.07)

- 1.07*** (0.14)

38

Table 5 THE EFFECT OF FIRM SIZE

This table reports regression estimates of the performance of public firms between 1993 and 2001 based on firm size (assets). Firms with assets greater than the industry median level are classified as large firms. The dependent variable is the firm’s industry-adjusted return on assets. Independent variables include a dummy variable for whether the firm was managed by a highly paid CEO (defined as those CEOs whose pay is in the top 40% of industry adjusted pay in the prior year). This dummy variable estimates the effect of highly paid CEOs relative to poorly paid CEOs (those CEOs whose industry adjusted pay is in the lower 40% in the prior year). The regression also includes a control for the prior year’s ROA. These variables are all interacted with dummy variable indicating whether the performance was above the industry median (good prior performance) or below the industry median (poor prior performance). To ease exposition, we report the estimated coefficients in two columns. The sample includes all firms for which executive compensation, accounting and stock price data is available on Execucomp, Compustat and CRSP. Standard errors are in parentheses and are adjusted for clustering. Estimated coefficients are significant at 1(***), 5(**), and 10 (*) percent.

Above Industry Median Size Below Industry Median Size

PANEL A: USING 1 YEAR LAG BETWEEN PERFORMANCE MEASURES






-0.07 0.06

0.08 0.09

0.16** 0.08

-0.10 0.07

Prior performance 0.82*** 0.06

0.27*** 0.08

0.59*** 0.03

0.35*** 0.02

Highly paid CEO dummy 0.28 0.56

-0.47 0.79

-1.10 0.72

-0.08 0.88

Intercept 0.43 0.51

-0.32 0.89

0.90*** 0.33

-1.90*** 0.53

R-squared 0.34 0.27

N 3,271 2,537 PANEL B: USING 2 YEAR LAGS BETWEEN PERFORMANCE MEASURES






0.11 0.08

-0.38** 0.17

0.28*** 0.11

-0.05 0.08


0.42** 0.17

0.48*** 0.05

0.21*** 0.04

Highly paid CEO dummy -0.62 0.68

-0.58 1.01

-1.51* 0.78

-0.87 1.17

Intercept 1.68*** 0.64

-0.32 0.97

1.17*** 0.38

-1.14** 0.55

R-squared 0.25 0.19

N 2,973 1,549

39

Table 6 CEO PAY AND SKILL: USING ANNUAL ABNORMAL EQUITY RETURNS

This table reports regression estimates of the performance of public firms between 1990 and 2001. The dependent variable is the firm’s equity return corrected for risk and/or style effects using the four factor Fama-French model (see details in text) and adjusted for average industry return. Independent variables include a dummy variable for whether the firm was managed by a highly paid CEO (defined as those CEOs whose pay is in the top 40% of industry adjusted pay in the prior year). This dummy variable estimates the effect of highly paid CEOs relative to poorly paid CEOs (those CEOs whose industry adjusted pay is in the lower 40% in the prior year). The regression also includes a control for the prior year’s equity return. These variables are all interacted with dummy variable indicating whether the performance was above the industry median (good prior performance) or below the industry median (poor prior performance). To ease exposition, we report the estimated coefficients in two columns. The sample includes all firms for which executive compensation, accounting and stock price data is available on Execucomp, Compustat and CRSP. Standard errors are in parentheses and are adjusted for clustering. Estimated coefficients are significant at 1(***), 5(**), and 10 (*) percent.

All firms










0.11** (0.05)

0.00 (0.09)

0.22** 0.10

-0.25* 0.15

0.13 0.10

-0.13 0.19

Prior performance -0.06* (0.04)

0.09 (0.07)

-0.17* 0.09

0.37** 0.15

-0.05 0.04

0.01 0.08

Highly paid CEO dummy -0.15*** (0.05)

-0.12** (0.06)

-0.31*** 0.11

-0.41*** 0.18

-0.14 0.09

-0.21 0.17

Intercept -0.08* (0.04)

0.00 (0.10)

0.08 0.09

0.33 0.24

-0.10** 0.05

-0.08 0.06

R-squared 0.002 0.006 0.001

N 6,375 3,248 3,112

40

TABLE 7 INDUSTRY DECLINES AND CEO SKILL

We report the marginal impact of a highly paid CEO on firm performance when the industry experiences an overall decline in profitability (defined as a decline in the median firm ROA). We first select firms operating in industries where median ROA has declined from the prior year. We next partition firms by the fraction of CEO pay that is performance-based pay, dividing the sample into two subgroups, above median performance-based pay and below median firms. The table reports results from a regression estimating the firm’s industry-adjusted accounting performance. Independent variables include a dummy variable for whether the firm was managed by a highly paid CEOs (defined as those CEOs whose pay is in the top 40% of industry adjusted pay in the prior year). This dummy variable estimates the effect of highly paid CEOs relative to poorly paid CEOs (those CEOs whose industry adjusted pay is in the lower 40% in the prior year). The regression also includes a control for the prior year’s ROA. These variables are all interacted with dummy variable indicating whether the performance was above the industry median (good prior performance) or below the industry median (poor prior performance). To ease exposition, we report the estimated coefficients in two columns.. The sample includes all firms for which executive compensation, accounting and stock price data is available on Execucomp, Compustat and CRSP. Estimated coefficients are significant at 1(***), 5(**), and 10 (*) percent.

Above median Performance-Based Pay







0.65*** 0.08

-0.39*** 0.09

-0.02 0.09

-0.18*** 0.06


0.85*** 0.12

0.98*** 0.04

0.65*** 0.04

Highly paid CEO dummy 0.13 0.67

1.86** 0.76

0.80 1.16

0.77 0.61

Intercept -0.92 0.62

1.66** 0.71

-1.38 0.30

3.43*** 0.27

R-squared N

41

TABLE 8 CEO PAY, SKILL AND TURNOVER

This table examines firm performance (ROA) in a two year window surrounding CEO turnover and reports the correlation between performance under one CEO and the performance under the successor. Panel A reports the overall performance persistence, conditional on the change in pay between the old CEO and the new CEO, for firm ROA and abnormal return quintiles. Panel B breaks these numbers down to reflect performance persistence for those changes associated with pay increases and those associated with pay decreases. The sample includes all firms for which executive compensation, accounting and stock price data is available on Execucomp, Compustat and CRSP. T-statistics are in parentheses.

No pay change 1 quintile pay change 2 quintile pay change

Panel A 0.55 (12.21)

0.61 (13.89)

0.00 (0.58)

Pay decrease

Pay increase

Pay decrease

Pay increase

Panel B 0.55 (12.21)

0.61 (11.25)

0.60 (9.16)

0.21 (2.37)

0.00 (0.47)

N 287 135 150 56 108

42

TABLE 9

CEO PAY, SKILL AND TURNOVER This table examines regressions estimating firm performance (defined as industry adjusted return on assets) in a two year window surrounding CEO turnover and reports the correlation between performance under one CEO and the performance under the successor. Firms are first partitioned into above median performance-based pay (those where the fraction of CEO pay tied to stock price is above the industry median) or below the industry median performance-based pay. Independent variables include a dummy variable for whether the firm was managed by a highly paid CEO (defined as a CEOs whose pay is in the top 40% of industry adjusted pay in the prior year). This dummy variable estimates the effect of highly paid CEOs relative to poorly paid CEOs (those CEOs whose industry adjusted pay is in the lower 40% in the prior year). The regression also includes a control for the prior year’s ROA. These variables are all interacted with dummy variable indicating whether the performance was above the industry median (good prior performance) or below the industry median (poor prior performance). To ease exposition, we report the estimated coefficients in two columns. The sample includes all firms for which executive compensation, accounting and stock price data is available on Execucomp, Compustat and CRSP. Standard errors are in parentheses and are adjusted for clustering. Estimated coefficients are significant at 1(***), 5(**), and 10 (*) percent.







Prior performance * dummy for increase in pay

0.12 (0.17)

-0.46*** (0.07)

0.10 (0.16)

-0.03 (0.32)

Prior performance 0.85*** (0.09)

0.46*** (0.07)

0.85*** (0.08)

0.05 (0.05)

Dummy for increase in pay -0.19 (1.43)

-3.20 (1.93)

-0.02 (1.72)

0.63 (3.11)

Intercept 0.39 (0.95)

0.83 (1.33)

0.07 (0.76)

1.33 (1.06)

R-squared 0.35 0.42

N 459 275

43

TABLE 10 MANAGERIAL SKILL AND PAY DISPERSION

We report association between a highly paid CEO and firm performance. Firms are first partitioned into above and below median pay dispersion, where pay dispersion is defined as the difference between the 90th percentile and the 10th percentile industry total CEO pay. The table reports results from a regression estimating the firm’s industry-adjusted accounting performance. We report only the estimated coefficients for firms lead by highly paid CEOs conditional on positive prior performance (above industry median) and prior bad (below industry median) performance (or “reversals”), that estimates the effect of highly paid CEOs relative to poorly paid CEOs. The regression model includes other unreported independent variables, including a constant, a dummy for highly paid CEOs, and controls for prior year performance, and interaction terms. The sample includes all firms for which executive compensation, accounting and stock price data is available on Execucomp, Compustat and CRSP. Estimated coefficients are significant at 1(***), 5(**), and 10 (*) percent.

Below Median Pay Dispersion

Above Median Pay Dispersion






-0.06 (0.06)

0.05 (0.08)

0.13** (0.06)

- 0.03 (0.03)

Prior performance 1.00*** (0.12)

0.67** (0.35)

0.82*** (0.04)

0.27*** (0.4)

Highly paid CEO dummy 0.08 (0.40)

0.43 (0.63)

- 0.33 (0.64)

0.55 (1.04)

Intercept - 0.29 (0.27)

- 1.09*** (0.4)

1.60*** (0.48)

- 2.11*** 0.77)

R-squared .34 .29

N 2,412 2,980

the good the bad and the lucky ceo pay and skillweb-docs.stern.nyu.edu/clb/docs/06-005.pdfharvard...

Documents